Polymorphisms in the TOX3/LOC643714 Locus and Risk of Breast Cancer in African-American Women

The TOX3/LOC643714 locus on chromosome 16 was one of the first breast cancer regions to be identified through genome-wide association studies in populations of European and East Asian origin (1). Out of several single nucleotide polymorphisms (SNP) associated with the risk of breast cancer, the rs3803662 (a C-to-T transition) was the most strongly correlated with disease; each copy of the T allele of the rs3803662 SNP was associated with a 20% increase in the risk of breast cancer. Subsequent studies, also in European ancestry populations, showed that the risk conferred by the rs3803662 polymorphism was either restricted to or more strongly associated with estrogen receptor (ER)–positive tumors compared with ER-negative cancers (2, 3). Because most of the replication studies have been carried out in populations of European ancestry (2-4), it is unclear whether the same SNP is associated with risk of breast cancer in populations of African origin. In a subgroup of African American women (422 cases and 447 controls) from the Multiethnic Cohort study, the T-rs3803662 allele was associated with a lower risk of breast cancer, the opposite direction from the results in the other ethnic groups (2). In addition, a recent analysis of African American women (810 cases and 1,784 controls) from the Southern Community Cohort Study and the Nashville Breast Health Study, found no significant association between rs3803662 and seven other SNPs in the TOX3 gene and risk of breast cancer (5).

The reasons for the differences in results between African American women and women of European or Asian ancestry remain to be determined. The high allele frequency of the T-rs3803662 allele in African Americans (∼50%) as well as the considerable sample size of the Southern Community Cohort Study/Nashville Breast Health Study (810 cases and 1,784 controls) makes low statistical power an unlikely explanation. Difference in the linkage disequilibrium (LD) structure in the TOX3/LOC643714 region between populations of European and African origin is a more likely reason (Fig. 1). In the HapMap CEU population, the rs3803662 SNP resides in a LD block that is more than 80 kb and covers part of the TOX3 and LOC643714 loci. In the HapMap YRI population, this big LD block was split into smaller blocks and included some gaps of low LD. In particular, the rs3803662 SNP is located inside a small 4 kb LD block in HapMap Yoruba samples. It is therefore possible that the causal variant(s) tagged by rs3803662 SNP in populations of European ancestry will be tagged by different SNPs in populations of African ancestry. If we assume the same causal polymorphisms in Europeans and Africans, the smaller LD blocks observed in the latter allows a more accurate mapping of the causal variants.

We conducted a nested case-control study to narrow the position of the causal variant for breast cancer in the TOX3/LOC643714 region. To this end, we performed fine-scale mapping of the entire TOX3/LOC643714 locus including genotyping of the index rs3803662 SNP.

We conducted a nested case-control study within the ongoing Black Women's Health Study (BWHS) that has been described elsewhere (6). Briefly, the study began in 1995 when women 21 to 69 years of age from across the United States completed a 14-page postal health questionnaire. The initial cohort comprises 59,000 women who self-identified as “black” and had a valid address. Follow-up questionnaires are sent every 2 years. Average follow-up of the baseline cohort through the completed 2-year cycles to date is greater than 80%.

We used medical records and cancer registry data to confirm self-reported cases of breast cancer, as well as to gain information on tumor characteristics such as estrogen and progesterone receptor (PR) status. We have obtained records or registry data for 1,151 breast cancer cases reported on the BWHS questionnaires, of which 99.4% were confirmed. Self-reported cases that were not confirmed have been excluded.

DNA samples were obtained from BWHS participants by the mouthwash-swish method (7), with all samples stored in freezers at −80°C. Approximately 50% of participants (27,800 women) provided a sample. Women who provided samples were slightly older than women who did not, but the two groups were similar with regard to educational level, geographic region of residence, body mass index, and family history of breast cancer.

The present study includes all cases of breast cancer who provided a DNA sample and were diagnosed through the end of the 2007 follow-up cycle. Controls were selected from among BWHS participants with DNA samples who were free of breast cancer at the end of the 2007 follow-up period. Controls were matched to cases approximately 1:1 on year of birth (±1 year), and geographic region of residence (Northeast, South, Midwest, and West). The study protocol was approved by the Institutional Review Board of Boston University.

We downloaded SNPs covering the entire TOX3/LOC643714 locus from the HapMap Yoruba (YRI) database (8). We used the Tagger (9) software implemented in Haploview (10) version 4.1 to select the set of common haplotype-tagging SNPs with a minor allele frequency of ≥5% and r² ≥ 0.8. The rs3803662 SNP was forced into the set. We selected 68 tagging SNPs along the TOX3/LOC643714 locus.

We also selected 30 ancestral informative markers (AIM) to estimate and control for population stratification due to European admixture. The 30 AIMs were selected from a list of validated SNPs in which the top 30 AIMs had allele frequency differences between Africans and Europeans of at least 0.75 (11). We used a Bayesian approach, as implemented in the Admixmap software (12, 13), to estimate individual admixture proportions. Eighty-one controls included in this breast cancer study had also been genotyped for a set of 1,536 AIMs used for admixture mapping analyses in a case-control study of systemic lupus erythematosus in the BWHS. The correlation (r = 0.87) between the two measures of percentage of European admixture was highly significant (P < 0.0001), confirming the validity of our small set of AIMs. Because Admixmap requires the specification of the ancestral allele frequencies, we also used the Structure software version 2.2 (14, 15), which does not require ancestral allele frequencies, to identify hidden population stratification beyond the one due to European admixture.

DNA was isolated from mouthwash-swish samples from breast cancer cases and controls at the Boston University Molecular Core Genetics Laboratory using the QIAAMP DNA Mini Kit (Qiagen). Whole genome amplification was done with the Qiagen RePLI-g Kits using the method of multiple displacement amplification. Amplified samples underwent purification and PicoGreen quantification at the Broad Institute Center for Genotyping and Analysis (Cambridge, MA) before being plated for genotyping.

Genotyping was carried out at the Broad Institute Center for Genotyping and Analysis using the Sequenom MassArray iPLEX technology. Ninety-eight blinded duplicate samples were included to assess the reproducibility of the genotypes. An average reproducibility of 99% was obtained among the blinded duplicates. All SNPs with a calling rate of <90% or a deviation from Hardy-Weinberg equilibrium in the control sample of P < 0.001 were excluded. We also excluded samples with calling rates of <80%. We first genotyped 41 tagging SNPs chosen for the TOX3 gene among 769 cases and 833 controls. After the addition of breast cancer cases and controls identified after the first genotyping plates were made, we genotyped 27 tagging SNPs along the LOC643714 locus in 865 cases and 1,073 controls. We successfully genotyped 35 of the TOX3 variants, 25 of the LOC643714 variants, and 29 of the AIMs. Mean call rate in the final data set for both SNPs and samples was 98.4%.

We used PLINK (16) software version 1.06 to calculate summary statistics for the genotype data. We tested for association with breast cancer using the Cochran-Armitage trend test of an additive genetic model with 10,000 permutations to calculate empirical P values. We used PROC LOGISTIC of the SAS statistical software version 9.1.3 (SAS Institute, Inc.) to estimate odds ratios (OR) and 95% confidence intervals (95% CI) for the SNPs significant at the nominal values (P = 0.05). We adjusted the ORs for age, geographic region of residence (Northeast, South, Midwest, West), place of birth (United States, foreign country), and European admixture proportion. We used a general genetic model with 2 df because it does not assume any particular inheritance risk pattern. Also, we estimated per allele ORs if a linear trend was evident.

We used the conditional haplotype method (17, 18) to determine whether the identified significant SNPs represent independent signals or were tagging the same causal variant. This method stratifies by haplotypic background and tests the null hypothesis that one or more SNPs have no independent haplotypic effect once we condition for such a background. We conditioned on the top SNP under the trend model to test for independent effects of the other significant SNPs.

ORs of haplotypes of significant SNPs were estimated using an expectation substitution approach (19, 20) that estimates the probabilities of all possible haplotype configurations of each individual in the sample, conditional on their genotype and case-control status. Haplotypes with an estimated frequency of <5% were pooled in one single group and the most common haplotype was used as the reference haplotype.

Table 1 shows the characteristics of breast cancer cases and controls. No significant differences were observed in the percentage of European admixture between the groups (19.2% in cases versus 19.3% in controls). The Admixmap and Structure software gave very similar results, with a correlation of 98.9% between the two estimates of European admixture proportions (Supplementary Fig. S1).

Because additional population stratification might be present beyond that due to European admixture, we estimated the likelihood of the observed AIM genotypes under different numbers of subpopulation groups. The observed data is best explained by two different subpopulations (African and European). However, our results also suggest the presence of a third or even a fourth subpopulation in the BWHS population (Supplementary Fig. S2). To assess the effect of this additional population stratification in our OR estimates, we conducted all logistic analyses under different scenarios of two, three, and four subpopulations. Because no major differences were observed among these three different scenarios, we present the results only under the scenario of two subpopulations.

The rs3803662 polymorphism was not significantly associated with risk of breast cancer overall (Table 2), or with particular subtypes of tumors defined by ER and PR status. Per allele ORs (95% CI) were 1.00 (0.81-1.25) for ER-positive tumors and 0.98 (0.76-1.26) for ER-negative breast cancer cases.

Of the SNPs scanned in the TOX3/LOC643714 region (Supplementary Table S1), four SNPs, all of them in the LOC643714 locus, were associated with breast cancer at the nominal α = 0.05 level of significance (Fig. 2). Two of the SNPs (rs3104746 and rs3112562) were located inside intron 2 of LOC643714, and the other two SNPs (rs3104793 and rs8046994) were located in the 5′ region of LOC643714 (Supplementary Fig. S3). The per allele ORs (95% CI) were 1.23 (1.05-1.44) for the A allele of rs3104746, 1.17 (1.02-1.34) for the G allele of rs3112562, 1.14 (1.00-1.30) for the C allele of rs3104793, and 1.16 (1.01-1.33) for the T allele of rs8046994 (Table 2). The similarity of the ORs for heterozygous and homozygous subjects suggests that risk of disease follows a dominant model. For all four SNPs, the risk of disease was stronger under the dominant model compared with the dose-response model, although comparison of the log likelihood of both genetic models did not allow us to discriminate between them (data not shown). We observed similar ORs for different subtypes of tumors defined by ER and PR status (data not shown).

These four SNPs tended to be correlated with each other as measured by the D′ and r² values, and correlation with the rs3803662 SNP was lower (Table 3), suggesting that the four SNPs may be tagging a single causal variant not tagged by the rs3803662 SNP in African Americans. The conditional haplotype method supports the notion of a single causal variant tagged by these four SNPs. After conditioning by the top SNP, rs3104746 SNP, no other SNP was significantly associated with disease: rs3112562 (P = 0.19), rs3104793 (P = 0.51), and rs8046994 (P = 0.50). Haplotypic ORs show haplotypes carrying the A allele of the rs3104746 SNP tended to be associated with risk of breast cancer (Table 4). In particular, the rs3104746-A/rs3112562-G/rs3104793-C/rs8046994-T haplotype was more frequent in cases compared with controls (15.8% versus 13.0%) and was associated with a 36% increase in the risk of breast cancer, OR (95% CI) = 1.36 (1.10-1.67). We note that the frequency of the A allele of the rs3104746 SNP in HapMap CEU samples is 4.2% compared with 25.4% in the HapMap Yoruba samples and 19.6% in the BWHS control population, and has an r² = 0.10 with rs3803662; therefore, the rs3104746 polymorphism may not be a good tagger of the causal variant in populations of European ancestry.

The rs3104746 and rs3112562 SNPs are in the same 87 kb LD block of the HapMap CEU sample that contains the rs3803662 SNP. The rs3104793 and rs8046994 SNPs are located in an adjacent 24 kb LD block in the same HapMap CEU sample. We used permutation analysis to evaluate the significance of our results adjusting for multiple comparisons within each LD block (27 SNPs in the large block and 10 SNPs in the small block). Both the trend and dominant models were assessed, with 100,000 permutations within each LD block for each of the genetic models. For the trend model, permutated P values were 0.20 for rs3104746, 0.30 for rs3112562, 0.23 for rs3104792, and 0.16 for rs8046994. For the dominant model, permutated P values were 0.16 for rs3104746, 0.12 for rs3112562, 0.05 for rs3104793, and 0.09 for rs8046994.

In the present study, we confirm the previous finding from genome-wide association studies of European ancestry populations of an association of breast cancer risk with a locus in the TOX3/LOC643714 region. The SNP associated with breast cancer in previous reports tagged a wide LD region. Our results point to a narrower region of association, located entirely within the LOC643714 gene.

The rs3803662 SNP, associated with breast cancer in European and Asian ancestry populations, was not associated with risk of breast cancer in the BWHS population. Furthermore, no evidence of association was found with any of the subtypes of breast cancer defined by ER and PR status. In African American women from the Multiethnic Cohort study, the T allele was associated with lower risk of breast cancer, an association opposite in direction to the results from other ethnic groups (2). However, in a recent study by Zheng et al. (5) in African American women of the Southern Community Cohort Study and the Nashville Breast Health Study, there was no significant association between the rs3803662 SNP and risk of breast cancer, in concordance with our present results. It is noteworthy that the study by Zheng et al. (5) also included seven SNPs in the TOX3 gene that are in high LD (r² ≥ 0.8) with rs3803662 in Europeans, and none of these polymorphisms was associated with risk of breast cancer. However, such a small number of SNPs is not enough to cover the genetic variation across the TOX3/LOC643714 region in African Americans. These two studies suggest that the causal variant in the TOX3/LOC643714 locus is not tagged by the rs3803662 SNP in African Americans. Our results suggest that causal variant(s) are not located in the TOX3 gene but rather in the LOC643714 locus. Consistent with our results, ongoing fine-mapping excluded the coding region of the TOX3 gene using 2,270 breast cancer cases and 2,280 controls in a European population from the United Kingdom and narrowed the associated region from the 5′ end of the TOX3 gene through the 3′ end of the LOC643714 locus (21). At the time of this writing, no function is known for the LOC643714 gene. According to Entrez Nucleotide and based on computational analysis, the LOC643714 locus codes for a small mRNA of 1,028 bp that would be translated into a hypothetical protein of 55 amino acids (22).

Although we used AIMs to estimate European admixture proportions, residual confounding due to population stratification might still be present. Even if we cannot completely rule out the presence of residual confounding, we think its effects, if any, on the present results are negligible. First, the selected AIMs provided us with estimates of European ancestry highly correlated with estimates using 1,536 AIMs distributed throughout the genome. Also, the present estimates of European admixture are similar to those reported from other African American populations (11, 23, 24). Control for European admixture removed the major source of confounding due to population stratification. Although more subtle population stratifications may still exist, the extra adjustment for a third or even a fourth subpopulation as identified by the Structure software did not materially change our OR estimates. We also note that we adjusted for geographic region of residence and birthplace, therefore reducing potential confounding due to population stratification.

In summary, our present results provide evidence that the TOX3/LOC643714 locus may contribute to the genetic susceptibility of breast cancer in African ancestry populations. The newly identified genetic variants located in the LOC643714 gene might be tagging the same causal variant. These findings help to narrow the localization of the causal variant(s) in the TOX3/LOC643714 region.

No potential conflicts of interest were disclosed.

We thank the Black Women's Health Study participants for their continuing participation in this research effort. We also thank Dr. Clint Baldwin of Boston University for advice on genotyping and interpretation of genotype data.

Grant Support: R01CA058420 and R01CA098663 from the National Cancer Institute, Division of Cancer Control and Population Science (http://www.cancercontrol.cancer.gov). The Broad Institute Center for Genotyping and Analysis is supported by grant U54 RR020278 from the National Center for Research Resources (http://www.broadinstitute.org/sections/science/projects/broad/ncrr-center-genotyping-analysis).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.